The characterization of sedimentary structures is an important step in constructing quantitative models of sedimentary deposits from digital images, such as 3D seismic data, satellite images, or digital outcrops. However, the interpretation of these structures generally consists of tedious line pickings followed by surface modeling to define geobodies. Automatic geobody extraction is an alternative, but it is sensitive to image noise, and it does not account for prior sedimentary knowledge. We decided to combine minimal picking by an interpreter with object-guided image processing and optimization to achieve fast and semiautomatic geobody interpretation. Our approach used a realistic volumetric geobody representation based on nonuniform rational basis splines, which can easily be deformed by the interpreter and numerical optimization. Custom edge detection guided by some initial rough interpretations was performed to strengthen the most relevant edges in the picture. Automatic optimization was then computed to fit the initial geobody to these highlighted edges. This approach was applied on satellite pictures showing alluvial channels, and some preliminary results on 3D seismic time slices were also presented. The interpreted channels were then used in a retrodeformation process to automatically reconstruct the point bars. This semiautomatic method opens new perspectives to help interpreters rapidly come up with 3D models of sedimentary structures from subsurface and analog surface data sets.